敏感性、特异性、假阳性、假阴性(sensitivity and specificity)

医学、机器学习等等,在统计结果时时长会用到这两个指标来说明数据的特性。

定义

敏感性:在金标准判断有病(阳性)人群中,检测出阳性的几率。真阳性。(检测出确实有病的能力)
特异性:在金标准判断无病(阴性)人群中,检测出阴性的几率。真阴性。(检测出确实没病的能力)
假阳性率:得到了阳性结果,但这个阳性结果是假的。即在金标准判断无病(阴性)人群中,检测出为阳性的几率。(没病,但却检测结果说有病),为误诊率。
假阴性率:得到了阴性结果,但这个阴性结果是假的。即在金标准判断有病(阳性)人群中,检测出为阴性的几率。(有病,但却检测结果说没病),为漏诊率。

计算方法

Sensitivity and specificity:完整定义

敏感性、特异性、假阳性、假阴性(sensitivity and specificity)

 

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
True Positive (真正, TP)被模型预测为正的正样本;可以称作判断为真的正确率
 
True Negative(真负 , TN)被模型预测为负的负样本 ;可以称作判断为假的正确率
 
False Positive (假正, FP)被模型预测为正的负样本;可以称作误报率
 
False Negative(假负 , FN)被模型预测为负的正样本;可以称作漏报率
 
True Positive Rate(真正率 , TPR)或灵敏度(sensitivity)
TPR = TP /(TP + FN)
正样本预测结果数 / 正样本实际数
 
True Negative Rate(真负率 , TNR)或特指度(specificity)
TNR = TN /(TN + FP)
负样本预测结果数 / 负样本实际数
 
False Positive Rate (假正率, FPR)
FPR = FP /(FP + TN)
被预测为正的负样本结果数 /负样本实际数
 
False Negative Rate(假负率 , FNR)
FNR = FN /(TP + FN)
被预测为负的正样本结果数 / 正样本实际数

  

假阳性率=假阳性人数÷金标准阴性人数

即: 假阳性率=b÷(b+d)

    金标准 金标准  
    阳性(+) 阴性(-) 合计
某筛检方法 阳性(+) a b a+b
某筛检方法 阴性(-) c d c+d
合计   a+c b+d N

公式为:假阳性率=b/(b+d)×100%

(b:筛选为阳性,而标准分类为阴性的例数;d:阴性一致例数)

假阴性率=假阴性人数÷金标准阳性人数

即: β=c÷(a+c)


终于要用到这个玩意了,很激动,主要统计假阴假阳性率。

我的任务:

1. 评估Pacbio MHC variation calling 结果(CCS/non-CCS)与Hiseq数据结果的一致性。
2. 分别在不同深度梯度的区域完成以上评估,推断PB MHC做variation calling的最低深度。

这里要将一个位点分为SNP、REF 和 LowQual,然后只去 SNP 和 REF 进行统计。

这是我一下午写出来的统计代码:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
#!/usr/bin/env python
# Author: LI ZHIXIN
 
import sys
import pysam
from collections import OrderedDict
 
def classify_DP(depth):
    if depth > 101:
        return 21
    return ((depth-1)//5+1)
 
def parse_rec(rec):
    sample = list(rec.samples)[0]
    # filter the Invalid line
    if not ('GQ' or 'GT' or 'DP') in rec.samples[sample].keys() or len(rec.alleles) <= 1:
        # continue
        return 1, "LowQual", rec.pos
    # filter the LowQual
    if rec.samples[sample]['GQ'] < 30:
        return rec.samples[sample]['DP'], "LowQual", rec.pos
    # filter the indel
    flag = 0
    for one in rec.alleles:
        if len(one) != len(rec.ref):
            flag = 1
    if flag == 1:
        return rec.samples[sample]['DP'], "LowQual", rec.pos
    if rec.samples[sample]['GT'] != (0, 0): # rec.qual > 30
        # variation_dict[rec.pos] = ["snp", rec.alleles]
        return rec.samples[sample]['DP'], "snp", rec.pos 
    elif rec.samples[sample]['GT'] == (0, 0):
        # variation_dict[rec.pos] = ["ref", rec.alleles]
        return rec.samples[sample]['DP'], "ref", rec.pos
 
def read_gvcf(gvcf_file_path):
    variation_dict = OrderedDict()
    for i in range(1,22):
        variation_dict[i] = {}
        for j in ('LowQual', 'snp', 'ref'):
            variation_dict[i][j] = []
    # pos_list = []
    gvcf_file = pysam.VariantFile(gvcf_file_path)
    for rec in gvcf_file.fetch('chr6',28477796,33448354):
        DP, pos_type, pos = parse_rec(rec)
        if DP < 1 or DP > 20:
            continue
        # DP = classify_DP(DP)
        variation_dict[DP][pos_type].append(pos)
        # print(pos, DP, pos_type)
    gvcf_file.close()
    # return variation_dict, pos_list
    return variation_dict
 
def read_hiseq_gvcf(gvcf_file_path):
    variation_dict = OrderedDict()
    # for i in range(1,22):
    # variation_dict[i] = {}
    for j in ('LowQual', 'snp', 'ref'):
        variation_dict[j] = []
    # pos_list = []
    gvcf_file = pysam.VariantFile(gvcf_file_path)
    for rec in gvcf_file.fetch('chr6',28477796,33448354):
        DP, pos_type, pos = parse_rec(rec)
        DP = classify_DP(DP)
        variation_dict[pos_type].append(pos)
        # print(pos, DP, pos_type)
    gvcf_file.close()
    # return variation_dict, pos_list
    return variation_dict
 
def show_dict_diff_DP(Hiseq_unified_variation_dict, PB_non_CCS_variation_dict, outf, outf2):
    for DP in range(1,21):
        Hiseq_snp = set(Hiseq_unified_variation_dict['snp'])
        Hiseq_ref = set(Hiseq_unified_variation_dict['ref'])
        Hiseq_lowqual = set(Hiseq_unified_variation_dict['LowQual'])
        PB_snp = PB_non_CCS_variation_dict[DP]['snp']
        PB_ref = PB_non_CCS_variation_dict[DP]['ref']
        PB_lowqual = PB_non_CCS_variation_dict[DP]['LowQual']
        total = set(PB_snp + PB_ref + PB_lowqual)
        Hiseq_snp = total & Hiseq_snp
        Hiseq_ref = total & Hiseq_ref
        Hiseq_lowqual = total & Hiseq_lowqual
        PB_snp = set(PB_snp)
        PB_ref = set(PB_ref)
        PB_lowqual = set(PB_lowqual)
        a = len(Hiseq_snp & PB_snp)
        b = len(Hiseq_ref & PB_snp)
        c = len(Hiseq_lowqual & PB_snp)
        d = len(Hiseq_snp & PB_ref)
        e = len(Hiseq_ref & PB_ref)
        f = len(Hiseq_lowqual & PB_ref)
        g = len(Hiseq_snp & PB_lowqual)
        h = len(Hiseq_ref & PB_lowqual)
        i = len(Hiseq_lowqual & PB_lowqual)
        Low_total = (g+h+i)/(a+b+c+d+e+f+g+h+i)
        if (a+b) == 0:
            PPV = "NA"
        else:
            PPV = a/(a+b)
            PPV = "%.4f"%(PPV)
        if (a+d) == 0:
            TPR = "NA"
        else:
            TPR = a/(a+d)
            TPR = "%.4f"%(TPR)
        print(str(DP)+" :\n", a,b,c,"\n",d,e,f,"\n",g,h,i,"\n", file=outf2, sep='\t', end='\n')
        print(DP, TPR, PPV, "%.4f"%Low_total, file=outf, sep='\t', end='\n')
 
with open("./depth_stat.txt", "w") as outf:
    print("Depth", "TPR", "PPV", "Low_total", file=outf, sep='\t', end='\n')
    outf2 = open("raw.txt", "w")
    Hiseq_unified_variation_dict = read_hiseq_gvcf("./hiseq_call_gvcf/MHC_Hiseq.unified.gvcf.gz")
    PB_non_CCS_variation_dict = read_gvcf("./non_CCS_PB_call_gvcf/MHC_non_CCS.unified.gvcf.gz")
    show_dict_diff_DP(Hiseq_unified_variation_dict, PB_non_CCS_variation_dict, outf, outf2)
    outf2

  

又碰到一个高级python语法:在双层循环中如何退出外层循环? 我用了一个手动的flag,有其他好方法吗?

如何统计下机数据的覆盖度和深度?当然要比对之后才能统计,而且还要对比对做一些处理。

在计算一个位点是否是SNP、indel、Ref时,不仅要考虑ref、alts、qual、GQ,而且必须要把GT、DP考虑在内,所以说还是比较复杂的。

 

最后如何分析第二个问题,call variation的最低深度?

统计不同深度下的假阴假阳性率,看在什么深度下其达到饱和。